基于结构特征对齐的γ辐射环境场景图像配准方法

γRadiation Scene Image Registration Method Based on Structural Feature Alignment

针对γ辐射环境场景图像中位置、形状、颜色多样的斑块噪声在图像配准过程中比场景信息更加显著,严重干扰场景信息的特征表达,导致场景图像难以实现准确图像对齐的问题,本文提出一种联合去噪策略与结构特征对齐的γ辐射环境场景图像配准方法,该方法从降低噪声干扰和改善场景信息的特征表达两个方面提升配准精度。首先,针对γ辐射环境场景图像中斑块噪声多样性的问题,通过金字塔滑动取样策略将γ辐射环境场景图像下采样为无损子图像集,使斑块噪声被拆分为子图像中的孤立点噪声,并利用中值滤波方法对此类噪声进行去除。其次,针对场景信息特征显著性不足的问题,采用Sobel边缘特征提取方法获取场景信息的稳定结构特征,实现对斑块噪声边缘弱纹理的抑制及场景信息特征表达的提升。最后,利用傅里叶-梅林变换方法进行结构特征的准确对齐,实现γ辐射环境场景图像的准确配准。在^(60)Coγ辐射环境采集的图像上进行了大量对比实验,待配准图像的尺度量解析误差小于0.01,旋转量解析误差小于0.4°,平移量解析误差小于±0.6像素,实验结果表明了本文方法的有效性。

Due to the impact of the high-energy photons on the vision sensor,the images captured from theγradiation environment are often polluted by the speckle noise with random location,disordered shapes,and uncertain color distribution.The uncertainty of the speckle noise leads to the lack of representation consistency of the texture features,brightness distribution,structural features,and other image descriptors for the radiation image,which seriously limits the feature expression of general feature descriptors for the key scene information.The limitation of representing the objects’features in the radiation scenes directly aggravates the difficulty of the image alignment task.In this paper,we proposed a novelγradiation scene image alignment approach jointed with a hierarchical noise suppression scheme and a sparse structure feature alignment strategy for the above-mentioned problems.The accuracy ofγimage alignment can be boosted with three strategies,including removing the salient noise areas,suppressing the fine-texture noise regions,and the sparse representation of object features.Concretely,for the problem of object’s feature affected by the high saliency of the central region of the speckle noise in the radiation scene images,we firstly utilized the pyramid sliding sampling strategy to down-sample theγradiation scene image into a set of lossless sub-images,where the speckle noise was divided into the isolated point noise in the sub-image.And then,a detection-based median-filtering scheme was presented to remove these isolated point noises.For the problem of insufficient saliency of feature representation for the scene objects in the image alignment task,a Sobel-based edge detection approach was used to extract the stable and sparse structure features of objects in the application scene,which can suppress the weak texture at the surrounding of the speckle noise and improve the robustness of scene object feature representation.Finally,the Fourier-Merlin transform,which is robust to different types of image noise,is utilized to accurately align the structural feature of the scene object to achieve the accurate alignment of theγradiation scene images.In order to prove the effectiveness and generalization of our proposed method,extensive experiments were conducted on theγradiation scene image set collected in the reality ^(60)Co radiation environment.With the help of our proposed method,we can achieve the alignment accuracy that the scale analysis error is less than 0.01,the rotation analysis error is less than 0.4°,and the translation analysis error is less than±0.6 pixel.Experimental results demonstrate the effectiveness and the generalization of our proposed method.